omh-ralplan-driver

Name: omh-ralplan-driver
Author: witt3rd

117stars9forks

How to install omh-ralplan-driver?

witt3rd/oh-my-hermes/omh-ralplan-driver

$npx -y skills add witt3rd/oh-my-hermes --skill omh-ralplan-driver

Installs into the current project.

Summary

Drive omh-ralplan: context package, rounds, distillation.

Files · 1

View on GitHub

SKILL.md

1	# OMH Ralplan Driver — driving the loop
2
3	Load this skill alongside `omh-ralplan` when you are the orchestrator
4	dispatching the loop. `omh-ralplan` is loaded by the Planner / Architect
5	/ Critic workers — it covers what they do inside each `delegate_task`
6	call. This skill covers what you do as the dispatcher: prep, dispatch,
7	distill, review.
8
9	The two skills have different readers and different jobs. Don't merge
10	them — worker context is precious; the dispatcher's playbook should
11	not ride into every subagent.
12
13	## Why this skill exists
14
15	Ralplan without disciplined orchestration converges on internal
16	consistency, not truth. The Planner / Architect / Critic triangle is
17	self-checking — it will produce a coherent stance about *whatever
18	you pointed it at*, including the wrong frame, the stale source, the
19	phantom constraint, and the question the user already settled.
20
21	The orchestrator is the only role with the vantage to:
22
23	- Carve principles before dispatching.
24	- Author a context package that licenses framing contests.
25	- Verify ground truth and adjacent mechanisms before subagents
26	inherit them as fact.
27	- Iterate the package with the user — half of any non-trivial
28	requirements come from the user's lived context, not from reading.
29	- Distill consensus into a canonical artifact stripped of
30	reviewer scaffolding.
31	- Apply a final quality gate before handing anything to the user.
32
33	Most of the failure modes named below are pre-dispatch failures. **Prep
34	is where quality is born.** Distillation and review preserve it. The
35	loop itself is just the well-instrumented engine in between.
36
37	## When to use ralplan vs just think
38
39	Use ralplan when the work has:
40
41	- Multiple legitimate decompositions of the problem.
42	- Load-bearing principles that need enforcing across subagents.
43	- Cross-cutting concerns the orchestrator might miss alone.
44	- A user who would otherwise be in a turn-by-turn proposal loop.
45
46	Don't use ralplan when the work is single-file, single-decision, or
47	obvious-to-solve. Overkill is its own failure mode. Use
48	`omh-deep-interview` first if the goal is ambiguous.
49
50	## The five-step playbook
51
52	### 1. Carve principles first (if they don't exist yet)
53
54	Ralplan without principles produces plausibly-reasoned nonsense. The
55	Planner/Architect/Critic triangle converges on internal consistency,
56	not on truth. Principles are the external anchor.
57
58	A good starting point: a small `PRINCIPLES.md` (≤15 entries) at the
59	project root. Each principle should be load-bearing and name the
60	failure it prevents. Include this file in every subagent's required
61	reading.
62
63	If principles don't exist yet, carve them first — even a quick draft
64	beats unprincipled ralplan. The file can iterate; the existence of
65	the anchor cannot.
66
67	### 2. Draft a context package — not just a prompt
68
69	The context package is a living document at
70	`docs/design/<domain>/context.md`. It is the single most load-bearing
71	artifact in the run. The Planner reads it. The Architect reads it.
72	The Critic reads it. A weak package produces a weak stance no matter
73	how good the subagents are.
74
75	Required sections:
76
77	- The design question. Single sentence, then a 2-3 paragraph
78	expansion. Name what the output must do, what it must honor, what
79	it is the foundation for.
80
81	- Artifact-type declaration. One sentence near the top: "This is
82	design-shaped, not requirements-shaped" — or vice versa. Forces
83	the orchestrator to commit before drafting dimensions, and tells
84	subagents which discipline rules apply (see Pitfall 15). Cheap to
85	add; surfaces the wrong-shape failure before dimensions get drafted
86	around the wrong frame.
87
88	- Sub-dimensions. A first cut of what the stance must address.
89	Not sacred — the META question below licenses the Critic to
90	rework them.
91
92	- Constraints. Principles, directives, committed prior decisions
93	the run must honor.
94
95	- Out of scope. What this run does NOT design.
96
97	- Required reading. Absolute paths only. Subagents cannot `cd`
98	and cannot work from summaries. Group by category: principles +
99	directives, inspiration / prior art, target platform, existing
100	code, and the context package itself.
101
102	- Contested questions. Six to eight places lazy consensus is
103	most likely. Seeds for the Critic. Not exhaustive.
104
105	- A META question. "Are these the right sub-dimensions?" This
106	licenses the Critic to contest the framing itself, which is the
107	single most load-bearing move a Critic can make. Without it the
108	Critic stays inside the frame and only checks compliance.
109
110	- What done looks like. Concrete output criteria. Include one
111	that forces "what is missing from the inspiration / what would
112	an ideal version have that the inspiration lacks?"
113
114	The orchestrator drafts the cont

People who liked this also viewed

6 picks

Type

paperclip-create-agentCreate new agents in Paperclip with governance-aware hiring. Use when you need to inspect adapter configuration options, compare existing agent configs, draft a new agent prompt/config, and submit a hSkillsJul 2026157k5
subagent-driven-developmentUse when executing implementation plans with independent tasks in the current sessionSkillsJul 2026154k259k
dispatching-parallel-agentsUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesSkillsJul 2026143k259k
browser-act-skill-forgeForges reusable Skill packages (SKILL.md + scripts) from website exploration via browser-act — no re-exploration later. Use when: user wants a reusable Skill for any website, needs to understand a sitSkillsJul 202685k4.6k
wizardGenerate an interactive bash wizard that walks a human through a manual procedure — third-party setup, a one-off migration, an A→B state transition — opening URLs, capturing values, confirming each stSkillsJul 202675k182k
wonda-cliUsing the Wonda CLI to generate images, videos, music, and audio from the terminal — plus LinkedIn, Reddit, and X/Twitter research and automationSkillsJul 202666k137

omh-ralplan-driver

117stars9forks

View on GitHub

How to install omh-ralplan-driver?

witt3rd/oh-my-hermes/omh-ralplan-driver

$npx -y skills add witt3rd/oh-my-hermes --skill omh-ralplan-driver

Installs into the current project.

Summary

Drive omh-ralplan: context package, rounds, distillation.

Files · 1

View on GitHub

SKILL.md

1	# OMH Ralplan Driver — driving the loop
2
3	Load this skill alongside `omh-ralplan` when you are the orchestrator
4	dispatching the loop. `omh-ralplan` is loaded by the Planner / Architect
5	/ Critic workers — it covers what they do inside each `delegate_task`
6	call. This skill covers what you do as the dispatcher: prep, dispatch,
7	distill, review.
8
9	The two skills have different readers and different jobs. Don't merge
10	them — worker context is precious; the dispatcher's playbook should
11	not ride into every subagent.
12
13	## Why this skill exists
14
15	Ralplan without disciplined orchestration converges on internal
16	consistency, not truth. The Planner / Architect / Critic triangle is
17	self-checking — it will produce a coherent stance about *whatever
18	you pointed it at*, including the wrong frame, the stale source, the
19	phantom constraint, and the question the user already settled.
20
21	The orchestrator is the only role with the vantage to:
22
23	- Carve principles before dispatching.
24	- Author a context package that licenses framing contests.
25	- Verify ground truth and adjacent mechanisms before subagents
26	inherit them as fact.
27	- Iterate the package with the user — half of any non-trivial
28	requirements come from the user's lived context, not from reading.
29	- Distill consensus into a canonical artifact stripped of
30	reviewer scaffolding.
31	- Apply a final quality gate before handing anything to the user.
32
33	Most of the failure modes named below are pre-dispatch failures. **Prep
34	is where quality is born.** Distillation and review preserve it. The
35	loop itself is just the well-instrumented engine in between.
36
37	## When to use ralplan vs just think
38
39	Use ralplan when the work has:
40
41	- Multiple legitimate decompositions of the problem.
42	- Load-bearing principles that need enforcing across subagents.
43	- Cross-cutting concerns the orchestrator might miss alone.
44	- A user who would otherwise be in a turn-by-turn proposal loop.
45
46	Don't use ralplan when the work is single-file, single-decision, or
47	obvious-to-solve. Overkill is its own failure mode. Use
48	`omh-deep-interview` first if the goal is ambiguous.
49
50	## The five-step playbook
51
52	### 1. Carve principles first (if they don't exist yet)
53
54	Ralplan without principles produces plausibly-reasoned nonsense. The
55	Planner/Architect/Critic triangle converges on internal consistency,
56	not on truth. Principles are the external anchor.
57
58	A good starting point: a small `PRINCIPLES.md` (≤15 entries) at the
59	project root. Each principle should be load-bearing and name the
60	failure it prevents. Include this file in every subagent's required
61	reading.
62
63	If principles don't exist yet, carve them first — even a quick draft
64	beats unprincipled ralplan. The file can iterate; the existence of
65	the anchor cannot.
66
67	### 2. Draft a context package — not just a prompt
68
69	The context package is a living document at
70	`docs/design/<domain>/context.md`. It is the single most load-bearing
71	artifact in the run. The Planner reads it. The Architect reads it.
72	The Critic reads it. A weak package produces a weak stance no matter
73	how good the subagents are.
74
75	Required sections:
76
77	- The design question. Single sentence, then a 2-3 paragraph
78	expansion. Name what the output must do, what it must honor, what
79	it is the foundation for.
80
81	- Artifact-type declaration. One sentence near the top: "This is
82	design-shaped, not requirements-shaped" — or vice versa. Forces
83	the orchestrator to commit before drafting dimensions, and tells
84	subagents which discipline rules apply (see Pitfall 15). Cheap to
85	add; surfaces the wrong-shape failure before dimensions get drafted
86	around the wrong frame.
87
88	- Sub-dimensions. A first cut of what the stance must address.
89	Not sacred — the META question below licenses the Critic to
90	rework them.
91
92	- Constraints. Principles, directives, committed prior decisions
93	the run must honor.
94
95	- Out of scope. What this run does NOT design.
96
97	- Required reading. Absolute paths only. Subagents cannot `cd`
98	and cannot work from summaries. Group by category: principles +
99	directives, inspiration / prior art, target platform, existing
100	code, and the context package itself.
101
102	- Contested questions. Six to eight places lazy consensus is
103	most likely. Seeds for the Critic. Not exhaustive.
104
105	- A META question. "Are these the right sub-dimensions?" This
106	licenses the Critic to contest the framing itself, which is the
107	single most load-bearing move a Critic can make. Without it the
108	Critic stays inside the frame and only checks compliance.
109
110	- What done looks like. Concrete output criteria. Include one
111	that forces "what is missing from the inspiration / what would
112	an ideal version have that the inspiration lacks?"
113
114	The orchestrator drafts the cont

People who liked this also viewed

6 picks

Type